Democratizing LLMs: An Exploration of Cost-Performance Trade-offs in Self-Refined Open-Source Models