Is it possible to parallelize multiple ROI's with CUDA?

edit

I would like to know if OpenCV + CUDA can provide a solution to my problem.

I have an image with many (hundreds) small Region of Interest (say, each one is about 50-by-50 pixels). These ROI's are of different sizes. Supposedly, each ROI contains one object. I need to compute various features for each ROI, e.g. the major axis and minor axis of the object.

Is there a way to do this with OpenCV?

My concern is that many OpenCV GPU modules take one input image, divide it into blocks and apply single-instruction multiple-data operations to the blocks. But what I need is to operate on the different ROI's in parallel.

Thanks a lot!