The integration of System Dynamics (SD) and Machine Learning (ML) is increasingly necessary for modeling complex systems, yet current proprietary SD tools and transpiler-based open-source libraries present severe architectural bottlenecks that hinder deep algorithmic coupling. This thesis introduces OpenCLD, a Python-native, open-source library designed to resolve this structural disconnect by implementing a strict ``Model-as-Code'' paradigm. Built on an object-oriented architecture, the framework treats SD components as mutable, explicit Python objects. The simulation engine is optimized for high-performance execution, featuring O(1) component lookups, automated topological sorting via NetworkX to resolve dependencies and algebraic loops, and a robust numerical perturbation algorithm for dynamic polarity detection.