Explaining Deep Learning Models -- A Bayesian Non-parametric Approach