Optimization of Finite-Differencing Kernels for Numerical Relativity Applications