In this paper, a new relaxation algorithm for solving geometric constraint-based models is proposed. The algorithm starts from a constructive symbolic representation of objects (Constructive Parametric Solid Model, CPSM) and proceeds by iterative relaxation of the geometric constraints. Models that can be reduced to distance and angle constraints can be handled. A new algorithm based on an iterative global deformation of the system is presented and discussed, and its convergence is proved. The performance of hybrid algorithms involving global deformation and individual constraint relaxation is discussed on several practical cases.