Improving Performance of Iterative Applications through Interleaved Execution of Approximated CUDA Kernels