closer the distance, proximity for the similarity of class is more. In order to minimize the knn compute time. we have designed opencl base parallel algorithm for many core highly parallel architectures. the results for our opencl base parkmn proves that the performance scales up sub-linearly towards the improved performance of parallel knn. the classification accuracy is not identified with the proportional value of k.the better k is to be identified by multiple observations only. however,concerned to performance our parallel knn, parknn, we observed satisfactory scale up