We present a method to automatically deploy the peripheral section of the cardiac conduction system in ventricles. The method encodes anatomical information thorough rules that ensure that Purkinje network structures generated are realistic and comparable to those observed in ex-vivo studies. The core methodology is based in non-deterministic production rules that are parameterized by means of statistical functions. Input parameters allow the construction of a great diversity of Purkinje structures that could be incorporated in fine element ventricular models to perform electrophysiology simulations. Resulting Purkinje trees show good geometrical approximations of Purkinje core network and bundles when compared to histological diagrams and do not require user interaction. Simulations carried out with these models result in activation sequences remarkably similar to micro-electrode electrical mapping studies.