diff --git a/CMakeLists.txt b/CMakeLists.txt
index 963313c7f7a5821bdec123e658a07aa6233b28a9..12dcd69019fc3901d1aaff74af9957ad05817145 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -45,14 +45,14 @@ file(MAKE_DIRECTORY ${GEN_INTERFACE_DIR})
 #================================================================================
 set(RUNTIME_COMMON_FILES
     runtime/pairs.cpp
-    runtime/copper_fcc_lattice.cpp
-    runtime/create_body.cpp
-    runtime/dem_sc_grid.cpp
-    runtime/read_from_file.cpp
-    runtime/stats.cpp
-    runtime/thermo.cpp
-    runtime/timing.cpp
-    runtime/vtk.cpp
+    runtime/utility/copper_fcc_lattice.cpp
+    runtime/utility/create_body.cpp
+    runtime/utility/dem_sc_grid.cpp
+    runtime/utility/read_from_file.cpp
+    runtime/utility/stats.cpp
+    runtime/utility/thermo.cpp
+    runtime/utility/timing.cpp
+    runtime/utility/vtk.cpp
     runtime/domain/regular_6d_stencil.cpp)
 
 #================================================================================
@@ -106,9 +106,9 @@ if(USE_WALBERLA)
 
     # TODO: Generate the host/device functions for computing weights
     if(COMPILE_CUDA)
-        list(APPEND RUNTIME_WALBERLA_FILES runtime/boundary_weights.cu)
+        list(APPEND RUNTIME_WALBERLA_FILES runtime/domain/boundary_weights.cu)
     else()
-        list(APPEND RUNTIME_WALBERLA_FILES runtime/boundary_weights.cpp)
+        list(APPEND RUNTIME_WALBERLA_FILES runtime/domain/boundary_weights.cpp)
     endif()
     
     target_sources(${PAIRS_TARGET} PRIVATE ${RUNTIME_WALBERLA_FILES})
diff --git a/README.md b/README.md
index 92db3b5b5c1cab339f890dc54a874427232580b5..4d62ccb2b2b5e6d15b93875ac311a3ebc0597994 100644
--- a/README.md
+++ b/README.md
@@ -109,18 +109,17 @@ In the following, we assume we have created and navigated to a build directory:
 **Basic CMake flags:**  
 * Pass your input script to CMake using `-DINPUT_SCRIPT=path/to/script.py`  
 * Enable CUDA with `-DCOMPILE_CUDA=ON`
-* Enable support for BlockForest domain partitioning and dynamic load balancing by providing the path to waLBerla source directory `-DWALBERLA_DIR=path/to/walberla` (TODO: waLBerla as a submodule)
+* Enable waLBerla support with `-DUSE_WALBERLA=ON` for using BlockForest domain partitioning and dynamic load balancing
 
 
 ### 1. Stand-Alone P4IRS Application
 ---------------------
 To build a C++ application using P4IRS, provide the list of your source files to CMake using the `-DUSER_SOURCE_FILES` flag (semicolon-seperated).
 
-**Example**: Build the application [sd_1.cpp](examples/modular/sd_1.cpp) using [spring_dashpot.py](examples/modular/spring_dashpot.py) as the input script.  
-Note: In this example we assume waLBerla has been already cloned next to the P4IRS directory.
+**Example**: Build the application [sd_1.cpp](examples/modular/sd_1.cpp) using [spring_dashpot.py](examples/modular/spring_dashpot.py) as the input script.
 
 ```
-cmake -DINPUT_SCRIPT=../examples/modular/spring_dashpot.py -DUSER_SOURCE_FILES=../examples/modular/sd_1.cpp -DWALBERLA_DIR=../../walberla ..
+cmake -DINPUT_SCRIPT=../examples/modular/spring_dashpot.py -DUSER_SOURCE_FILES=../examples/modular/sd_1.cpp -DUSE_WALBERLA=ON ..
 ```
 Now call `make` and an **executable** is built.
 
diff --git a/runtime/domain/block_forest.cpp b/runtime/domain/block_forest.cpp
index 77c082b709787e96ea3fb4da923f06ef060f4ddc..1ad9922658bd11ca9f90731b68198810bcb674c6 100644
--- a/runtime/domain/block_forest.cpp
+++ b/runtime/domain/block_forest.cpp
@@ -13,7 +13,7 @@
 #include <blockforest/loadbalancing/weight_assignment/MetisAssignmentFunctor.h>
 #include <blockforest/loadbalancing/weight_assignment/WeightAssignmentFunctor.h>
 //---
-#include "../boundary_weights.hpp"
+#include "boundary_weights.hpp"
 #include "../pairs_common.hpp"
 #include "../devices/device.hpp"
 #include "regular_6d_stencil.hpp"
diff --git a/runtime/boundary_weights.cpp b/runtime/domain/boundary_weights.cpp
similarity index 100%
rename from runtime/boundary_weights.cpp
rename to runtime/domain/boundary_weights.cpp
diff --git a/runtime/boundary_weights.cu b/runtime/domain/boundary_weights.cu
similarity index 100%
rename from runtime/boundary_weights.cu
rename to runtime/domain/boundary_weights.cu
diff --git a/runtime/boundary_weights.hpp b/runtime/domain/boundary_weights.hpp
similarity index 86%
rename from runtime/boundary_weights.hpp
rename to runtime/domain/boundary_weights.hpp
index e84348a0c8438255ca0090b765b56fd95f7deb10..3b7fd22e4132b243531e07ed885bc1883b229820 100644
--- a/runtime/boundary_weights.hpp
+++ b/runtime/domain/boundary_weights.hpp
@@ -6,8 +6,8 @@
 #include <fstream>
 #include <sstream>
 //---
-#include "pairs.hpp"
-#include "pairs_common.hpp"
+#include "../pairs.hpp"
+#include "../pairs_common.hpp"
 
 
 namespace pairs {
diff --git a/runtime/copper_fcc_lattice.cpp b/runtime/utility/copper_fcc_lattice.cpp
similarity index 100%
rename from runtime/copper_fcc_lattice.cpp
rename to runtime/utility/copper_fcc_lattice.cpp
diff --git a/runtime/copper_fcc_lattice.hpp b/runtime/utility/copper_fcc_lattice.hpp
similarity index 100%
rename from runtime/copper_fcc_lattice.hpp
rename to runtime/utility/copper_fcc_lattice.hpp
diff --git a/runtime/create_body.cpp b/runtime/utility/create_body.cpp
similarity index 100%
rename from runtime/create_body.cpp
rename to runtime/utility/create_body.cpp
diff --git a/runtime/create_body.hpp b/runtime/utility/create_body.hpp
similarity index 100%
rename from runtime/create_body.hpp
rename to runtime/utility/create_body.hpp
diff --git a/runtime/dem_sc_grid.cpp b/runtime/utility/dem_sc_grid.cpp
similarity index 100%
rename from runtime/dem_sc_grid.cpp
rename to runtime/utility/dem_sc_grid.cpp
diff --git a/runtime/dem_sc_grid.hpp b/runtime/utility/dem_sc_grid.hpp
similarity index 100%
rename from runtime/dem_sc_grid.hpp
rename to runtime/utility/dem_sc_grid.hpp
diff --git a/runtime/read_from_file.cpp b/runtime/utility/read_from_file.cpp
similarity index 100%
rename from runtime/read_from_file.cpp
rename to runtime/utility/read_from_file.cpp
diff --git a/runtime/read_from_file.hpp b/runtime/utility/read_from_file.hpp
similarity index 100%
rename from runtime/read_from_file.hpp
rename to runtime/utility/read_from_file.hpp
diff --git a/runtime/stats.cpp b/runtime/utility/stats.cpp
similarity index 100%
rename from runtime/stats.cpp
rename to runtime/utility/stats.cpp
diff --git a/runtime/stats.hpp b/runtime/utility/stats.hpp
similarity index 100%
rename from runtime/stats.hpp
rename to runtime/utility/stats.hpp
diff --git a/runtime/thermo.cpp b/runtime/utility/thermo.cpp
similarity index 100%
rename from runtime/thermo.cpp
rename to runtime/utility/thermo.cpp
diff --git a/runtime/thermo.hpp b/runtime/utility/thermo.hpp
similarity index 100%
rename from runtime/thermo.hpp
rename to runtime/utility/thermo.hpp
diff --git a/runtime/timing.cpp b/runtime/utility/timing.cpp
similarity index 100%
rename from runtime/timing.cpp
rename to runtime/utility/timing.cpp
diff --git a/runtime/timing.hpp b/runtime/utility/timing.hpp
similarity index 100%
rename from runtime/timing.hpp
rename to runtime/utility/timing.hpp
diff --git a/runtime/vtk.cpp b/runtime/utility/vtk.cpp
similarity index 100%
rename from runtime/vtk.cpp
rename to runtime/utility/vtk.cpp
diff --git a/runtime/vtk.hpp b/runtime/utility/vtk.hpp
similarity index 100%
rename from runtime/vtk.hpp
rename to runtime/utility/vtk.hpp
diff --git a/src/pairs/code_gen/cgen.py b/src/pairs/code_gen/cgen.py
index 13f4f8f9482e4a324e872bf623cd644c85c207ab..9a07c9444bb1a16d1c7628b3a3d6decd898d11af 100644
--- a/src/pairs/code_gen/cgen.py
+++ b/src/pairs/code_gen/cgen.py
@@ -154,15 +154,15 @@ class CGen:
         self.print("#include <stdlib.h>")
         self.print("//---")
         self.print("#include \"likwid-marker.h\"")
-        self.print("#include \"copper_fcc_lattice.hpp\"")
-        self.print("#include \"create_body.hpp\"")
-        self.print("#include \"dem_sc_grid.hpp\"")
         self.print("#include \"pairs.hpp\"")
-        self.print("#include \"read_from_file.hpp\"")
-        self.print("#include \"stats.hpp\"")
-        self.print("#include \"timing.hpp\"")
-        self.print("#include \"thermo.hpp\"")
-        self.print("#include \"vtk.hpp\"")
+        self.print("#include \"utility/copper_fcc_lattice.hpp\"")
+        self.print("#include \"utility/create_body.hpp\"")
+        self.print("#include \"utility/dem_sc_grid.hpp\"")
+        self.print("#include \"utility/read_from_file.hpp\"")
+        self.print("#include \"utility/stats.hpp\"")
+        self.print("#include \"utility/timing.hpp\"")
+        self.print("#include \"utility/thermo.hpp\"")
+        self.print("#include \"utility/vtk.hpp\"")
         self.print("")
         self.print("using namespace pairs;")
         self.print("")