Synthesis of sign language from a formal linguistic model using data-driven animation techniques is a challenging task. The process is either expensive and repetitive or produces utterances which lack comprehension. We introduce a layer-based approach to define a signing avatar for solving posture constraints and present how it can be used for a complete data-driven multi-track sign language synthesis system. Our method synthesizes a sign language description model by combining low-level linguistic constraints as well as pre-animated actions. To unify these techniques, we formalize our posture using separate layers, which gives us control over the low-level skeleton and mesh specification. Finally, we build this system on top of an open-source animation toolkit.