Is there any way to use nvbuff in cuda?

i used img data from libargus, transfromed img data in cuda, outputed data by NvDrmRenderer to HDMI.
data in cuda input was GPU memory,and output was data used by NvDrmRenderer was in cpu memory.
i used cudaHostAlloc to create a memory for sharing between gpu and cpu named as sharePtr.
In my program, it has to be copied (memcpy) between the sharePtr and Nvbuff.
is there any way to use nvbfuff in cuda kernel function to reduce the memcpy?
i work on nvidia nx.


Here is a similar topic.
Could you check if the suggestion can meet your requirement first?