Graph Neural Networks (GNNs) have emerged as powerful tools for processing graph-structured data, with applications spanning diverse domains. This study proposes and evaluates customized GNN architectures for graph classification, specifically targeting the recognition of American Sign Language (ASL). We leverage a customized ASL dataset, where gestures are represented as graph structures, nodes capture key hand landmarks, and edges model their spatial and temporal relationships. Several GNN architectures are designed and optimized with varying hyperparameters to enhance classification accuracy. These models are benchmarked against traditional deep learning approaches, such as Multi-layer Perceptrons (MLPs) and Convolutional Neural Networks (CNNs), which rely on non-graphical representations of the dataset. The study highlights the advantages of graph-based representations for ASL recognition, providing insights into the limitations of traditional grid-based approaches. These findings underscore the potential of GNN-based methods for real-world ASL translation tasks, while suggesting future directions for hyperparameter refinement and architectural optimization.