Bayesian multilayer perceptron

Good day! I implemented Bayesian neural network. All weights have normal distribution. I know that input data has different distributions for every feature ( normal, multinomial and soon). So my question is it a right idea to have conjunction distributions in the first layer.