diff --git a/modules/cudaarithm/src/arithm.cpp b/modules/cudaarithm/src/arithm.cpp
index b2107dd1f..08de4e428 100644
--- a/modules/cudaarithm/src/arithm.cpp
+++ b/modules/cudaarithm/src/arithm.cpp
@@ -315,13 +315,20 @@ void cv::cuda::dft(InputArray _src, OutputArray _dst, Size dft_size, int flags,
     // We don't support real-to-real transform
     CV_Assert( is_complex_input || is_complex_output );
 
-    GpuMat src_cont = src;
-
     // Make sure here we work with the continuous input,
     // as CUFFT can't handle gaps
-    createContinuous(src.rows, src.cols, src.type(), src_cont);
-    if (src_cont.data != src.data)
+    GpuMat src_cont;
+    if (src.isContinuous())
+    {
+        src_cont = src;
+    }
+    else
+    {
+        BufferPool pool(stream);
+        src_cont.allocator = pool.getAllocator();
+        createContinuous(src.rows, src.cols, src.type(), src_cont);
         src.copyTo(src_cont, stream);
+    }
 
     Size dft_size_opt = dft_size;
     if (is_1d_input && !is_row_dft)
diff --git a/modules/cudaarithm/src/cuda/integral.cu b/modules/cudaarithm/src/cuda/integral.cu
index db554eb30..4a70ab0de 100644
--- a/modules/cudaarithm/src/cuda/integral.cu
+++ b/modules/cudaarithm/src/cuda/integral.cu
@@ -50,51 +50,58 @@
 
 #include "opencv2/cudaarithm.hpp"
 #include "opencv2/cudev.hpp"
+#include "opencv2/core/private.cuda.hpp"
 
+using namespace cv;
+using namespace cv::cuda;
 using namespace cv::cudev;
 
 ////////////////////////////////////////////////////////////////////////
 // integral
 
-void cv::cuda::integral(InputArray _src, OutputArray _dst, GpuMat& buffer, Stream& stream)
+void cv::cuda::integral(InputArray _src, OutputArray _dst, Stream& stream)
 {
-    GpuMat src = _src.getGpuMat();
+    GpuMat src = getInputMat(_src, stream);
 
     CV_Assert( src.type() == CV_8UC1 );
 
-    GpuMat_<int>& res = (GpuMat_<int>&) buffer;
+    BufferPool pool(stream);
+    GpuMat_<int> res(src.size(), pool.getAllocator());
 
     gridIntegral(globPtr<uchar>(src), res, stream);
 
-    _dst.create(src.rows + 1, src.cols + 1, CV_32SC1);
-    GpuMat dst = _dst.getGpuMat();
+    GpuMat dst = getOutputMat(_dst, src.rows + 1, src.cols + 1, CV_32SC1, stream);
 
     dst.setTo(Scalar::all(0), stream);
 
     GpuMat inner = dst(Rect(1, 1, src.cols, src.rows));
     res.copyTo(inner, stream);
+
+    syncOutput(dst, _dst, stream);
 }
 
 //////////////////////////////////////////////////////////////////////////////
 // sqrIntegral
 
-void cv::cuda::sqrIntegral(InputArray _src, OutputArray _dst, GpuMat& buf, Stream& stream)
+void cv::cuda::sqrIntegral(InputArray _src, OutputArray _dst, Stream& stream)
 {
-    GpuMat src = _src.getGpuMat();
+    GpuMat src = getInputMat(_src, stream);
 
     CV_Assert( src.type() == CV_8UC1 );
 
-    GpuMat_<double>& res = (GpuMat_<double>&) buf;
+    BufferPool pool(Stream::Null());
+    GpuMat_<double> res(pool.getBuffer(src.size(), CV_64FC1));
 
     gridIntegral(sqr_(cvt_<int>(globPtr<uchar>(src))), res, stream);
 
-    _dst.create(src.rows + 1, src.cols + 1, CV_64FC1);
-    GpuMat dst = _dst.getGpuMat();
+    GpuMat dst = getOutputMat(_dst, src.rows + 1, src.cols + 1, CV_64FC1, stream);
 
     dst.setTo(Scalar::all(0), stream);
 
     GpuMat inner = dst(Rect(1, 1, src.cols, src.rows));
     res.copyTo(inner, stream);
+
+    syncOutput(dst, _dst, stream);
 }
 
 #endif