Autonomous vehicles could bring a significant benefit in terms of fuel consumption and emissions, besides safety and comfort, but the full potential can only be exploited if the appropriate (optimal) control is applied. However complexity of optimal control can prevent its use in practice. Therefore, we suggest to approximate the optimal vehicle control problem by a two-layer structure in which the upper level provides a solution for the reference speed trajectory using energy consumption as cost function, while the lower one optimally tracks the reference speed considering fuel consumption and emissions. The key advantage is the simplicity of the upper layer, which reduces the computational burden, albeit with some expected loss of optimality if compared to the global optimum. The control and results are shown using a model of a production vehicle with a CI engine. The possibility of using this approach to trade-off fuel vs. NOx is discussed. The authors believe that splitting the optimization problem into a multilayer structure is a sensible approach to enforce computational tractability while recovering part of the optimization potential.