Bayesian optimal designs for discriminating between non-Normal models