New optimal design criteria for regression models with asymmetric errors