Graph Neural Networks (GNNs) have become crucial tools for graph classification tasks, effectively modeling relational data. However, as the depth of GNNs increases, they often encounter the oversmoothing problem, where node embeddings converge to indistinguishable vectors, leading to performance degradation. While various strategies have been proposed to mitigate this issue, such as normalization and feature dropping, these approaches often fall short in fully addressing oversmoothing, as they may either reduce the model’s expressive power or fail to preserve important node features in deeper layers. In this paper, we introduce a novel GNN framework designed to maintain depth while avoiding oversmoothing. Our architecture integrates adaptive mechanisms, including recall operation, fusion techniques, and adaptive normalization processes, to enhance feature propagation and capture complex graph structures. Through extensive experiments on multiple datasets, our framework demonstrates superior performance, consistently outperforming existing methods, particularly in deep GNN configurations.