The rapid expansion of blockchain technologies has brought smart contracts to the forefront of decentralized applications. However, many of these contracts are prone to vulnerabilities and fraudulent behaviors, posing serious security risks. Deep learning methods, particularly Convolutional Neural Networks (CNNs), offer promising solutions for automated contract analysis. Despite their success, the design of effective CNN architectures remains a complex and manual process, often reliant on expert intuition and trial-and-error. In this paper, we present an automated approach to CNN architecture design tailored for smart contract classification. We formulate the architecture search as an optimization problem and employ an Evolutionary Algorithm (EA) to evolve a population of CNN architectures. Each candidate network is encoded as a binary graph, allowing the EA to explore a wide range of topologies and select the best-performing models based on classification accuracy. The selected architectures are further fine-tuned and evaluated on a dataset of labeled smart contracts. Our approach is benchmarked against manually designed CNNs and other automated design techniques, including reinforcement learning and evolutionary search methods. The results demonstrate that our method produces more compact and accurate models, effectively detecting vulnerabilities in smart contracts. This work contributes to advancing secure blockchain development by providing a scalable and data-driven solution for smart contract analysis.