diff --git a/CMakeLists.txt b/CMakeLists.txt
index d708a6b..c2a8fb7 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -108,14 +108,16 @@ IF ( (${CMAKE_C_COMPILER_ID} STREQUAL "GNU" OR ${CMAKE_C_COMPILER_ID} STREQUAL "
     SET (USE_CUDA ON)
     INCLUDE_DIRECTORIES(${CUDA_INCLUDE_DIRS})
     LINK_DIRECTORIES(${CUDA_TOOLKIT_ROOT_DIR}/lib64)
+    LINK_DIRECTORIES(${CUDA_TOOLKIT_ROOT_DIR}/lib64/stubs)
     
     MESSAGE (STATUS "cuda include: ${CUDA_INCLUDE_DIRS}")
     MESSAGE (STATUS "cuda libs: ${CUDA_TOOLKIT_ROOT_DIR}/lib64")
     MESSAGE (STATUS "cuda version: ${CUDA_VERSION}")
+    SET(CUDA_PROPAGATE_HOST_FLAGS OFF)
 
     SET (CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -lcudart -lcufft -lcublas -lnvToolsExt -DDKS_CUDA")
     SET (CUDA_NVCC_FLAGS "-arch=sm_35 -DDEBUG -lcufft -lcublas -lcudart -fmad=false")
-    
+    SET (CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS}  -DDEBUG -std=c++11 -D__wsu")    
     SET (CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} ${OPENCL_KERNELS}")
 
     #if cuda version >= 7.0 add runtime commpilation flags
@@ -140,6 +142,7 @@ IF ( (${CMAKE_C_COMPILER_ID} STREQUAL "GNU" OR ${CMAKE_C_COMPILER_ID} STREQUAL "
       MESSAGE(STATUS "OpenCL version : ${OpenCL_VERSION_STRING}")
       MESSAGE(STATUS "OpenCL include dir: ${OpenCL_INCLUDE_DIR}")
       MESSAGE(STATUS "OpenCL library dir: ${OpenCL_LIBRARY}")
+      SET(CMAKE_SKIP_RPATH TRUE)
       INCLUDE_DIRECTORIES(${OpenCL_INCLUDE_DIR})
       LINK_DIRECTORIES(${OpenCL_LIBRARY})
     ENDIF (OpenCL_FOUND)
diff --git a/cmake/DKSConfig.cmake.in b/cmake/DKSConfig.cmake.in
index d764963..9499d8e 100644
--- a/cmake/DKSConfig.cmake.in
+++ b/cmake/DKSConfig.cmake.in
@@ -1,4 +1,5 @@
-SET(${PROJECT_NAME}_CMAKE_CXX_FLAGS "${${PROJECT_NAME}_CXX_FLAGS}")
+SET(${PROJECT_NAME}_CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}")
 SET(${PROJECT_NAME}_INCLUDE_DIR "${CMAKE_INSTALL_PREFIX}/include")
 SET(${PROJECT_NAME}_LIBRARY_DIR "${CMAKE_INSTALL_PREFIX}/lib")
-SET(${PROJECT_NAME}_LIBRARY "dks")
\ No newline at end of file
+SET(${PROJECT_NAME}_LIBRARY "dks")
+SET(CMAKE_SKIP_RPATH ${CMAKE_SKIP_RPATH})
\ No newline at end of file
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index 47b1b69..64d0192 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -83,4 +83,4 @@ TARGET_LINK_LIBRARIES(testFFTSolverMIC dks ${Boost_LIBRARIES} ${CLFFT_LIBRARIES}
 #IF (NOT CUDA_VERSION VERSION_LESS "7.0")
   #ADD_EXECUTABLE(testChiSquareRT testChiSquareRT.cpp)
   #TARGET_LINK_LIBRARIES(testChiSquareRT dks)
-#ENDIF (NOT CUDA_VERSION VERSION_LESS "7.0")
\ No newline at end of file
+#ENDIF (NOT CUDA_VERSION VERSION_LESS "7.0")
diff --git a/test/testRandom.cpp b/test/testRandom.cpp
new file mode 100644
index 0000000..64ea0df
--- /dev/null
+++ b/test/testRandom.cpp
@@ -0,0 +1,81 @@
+#include <iostream>
+#include <string>
+#include <vector>
+#include <sys/time.h>
+
+#include "DKSBase.h"
+
+using namespace std;
+
+int main(int argc, char *argv[]) {
+
+  int size = 10;
+  bool apiSet = false;
+  char *api_name = new char[10];
+  char *device_name = new char[10];
+
+  for (int i = 1; i < argc; i++) {
+
+    if (argv[i] == string("-cuda")) {
+      strcpy(api_name, "Cuda");
+      strcpy(device_name, "-gpu");
+      apiSet = true;
+    }
+
+    if (argv[i] == string("-opencl")) {
+      strcpy(api_name, "OpenCL");
+      strcpy(device_name, "-gpu");
+      apiSet = true;
+    }
+
+    if (argv[i] == string("-N")) {
+      size = atoi(argv[i+1]);
+      i++;
+    }
+
+  }
+
+  if (!apiSet) {
+    strcpy(api_name, "Cuda");
+    strcpy(device_name, "-gpu");
+  }
+    
+  cout << "=========================BEGIN TEST=========================" << endl;
+  cout << "Use api: " << api_name << "\t" << device_name << endl;
+  cout << "Number of randoms: " << size << endl;
+
+  //init dks
+  int ierr;
+  DKSBase base;
+  base.setAPI(api_name, strlen(api_name));
+  base.setDevice(device_name, strlen(api_name));
+  base.initDevice();
+  base.callInitRandoms(size);
+
+  //create host vector to store results
+  double *host_data = new double[size];
+
+  //create device vector
+  void *device_data = base.allocateMemory<double>(size, ierr);
+  
+  for (int i = 0; i < 5; i++) {
+    //fill device vector with random values
+    base.callCreateRandomNumbers(device_data, size);
+
+    //read device vector
+    base.readData<double>(device_data, host_data, size);
+
+    //print host data
+    for (int i = 0; i < size; i++)
+      cout << host_data[i] << " ";
+    cout << endl;
+  }
+
+  //free device vector
+  base.freeMemory<double>(device_data, size);
+  
+  //free host data
+  delete[] host_data;
+
+  return 0;
+}