Due to the growing use of AI models in sensitive sectors, and the consequent demand to explain how they make their predictions, multiple algorithms have been developed in the Explainable AI (XAI) sector to try to explain or justify these predictions. However, the development and evaluation of these algorithms are often not standardized. Our goal is to consider different evaluation metrics of XAI algorithms to automatically synthesize new XAI algorithms that outperform widely used algorithms, like GradCAM.