This paper presents a novel approach for simulating 3D muscle deformations with complex architectures. The approach consists in choosing the best model formulation in terms of computation cost and accuracy, that mixes a volumetric tissue model based on finite element method (3D FEM), a muscle fiber model (Hill contractile 1D element) and a membrane model accounting for aponeurosis tissue (2D FEM). The separate models are mechanically binded using barycentric embeddings. Our approach allows the computation of several fiber directions in one coarse finite element, and thus, strongly decreases the required finite element resolution to predict muscle deformation during contraction. Using surface registration, fibers tracks of specific architecture can be transferred from a template to subject morphology, and then simulated. As a case study, three different architectures are simulated and compared to their equivalent one dimensional Hill wire model simulations