Bayesian Optimal Designs for Linear Regression Models