Improving Large Language Models Function Calling and Interpretability via Guided-Structured Templates