So does that mean that there are other kernels that approximate derivatives "better"? Like with finite differences?