Modeling compensation of nonlinear friction is vital to improve the trajectory tracking performance of high acceleration motion systems. To overcome the problem of inaccurate estimation of the start-stop stage nonlinear friction (including friction overshoot) associated with the traditional parametric model for high acceleration motion systems, this paper describes a novel extended parametric model combining the traditional model structure with the extended Stribeck model. The training data for identifying the model parameters are obtained using the high-precise Iterative Learning Control (ILC) approach, which supplies the nonlinear friction feed-forward compensation data with limited trajectories in the workspace. The data are fitted with the Levenberg-Marquardt algorithm. Finally, the proposed model is validated with different trajectories on a high acceleration position platform driven by a Voice Coil Motor (VCM). The experimental results indicate that the proposed method can overcome the influence of nonlinear friction associated with the traditional model, including the friction overshoot in the start-stop stage. Moreover, the accuracy is comparable with the result of ILC, but offers the advantage that the proposed model can avoid the problem of poor generalization in ILC to realize the friction compensation of an arbitrary trajectory in the workspace.