D-efficient Bayesian Designs for a Class of Nonlinear Models