Efficient modulation classification remains a key challenge in contemporary communication systems research. In this work, we introduce a novel framework rooted in the principle of divide and conquer to address the complexities associated with jointly classifying linear and nonlinear modulation schemes. Our approach decomposes the classification task into multiple binary problems, effectively leveraging a hierarchical tree-based structure that integrates several low-parameterized CNNs. Simulation results validate the efficacy of our proposed method, demonstrating superior classification performance with a notable reduction in complexity compared to conventional single CNN- based approaches.