With the rapid development of secure multi-party computation (MPC) over past decades, applications of MPC has been moving from completing simple computation tasks (e.g., private set intersection) to complex computation tasks (e.g., privacy-preserving machine learning). This is an inevitable trend when more strict privacy protection requirements face more complex and large-scale computation such as big data analytics being applied in many fields. Although the complex computation tasks are not easy to be evaluated with one type of MPC protocols from beginning to the end, it can be more efficiently evaluated by decomposing the complex task into many simple sub-tasks and evaluating each of them with the proper type of MPC protocol in sequence. Therefore, we propose a mixed-protocol MPC framework towards complex computation tasks with malicious security in this work. In particular, we utilize the homomorphic commitment technique to construct six types of share conversion protocols in the malicious model. Then, we construct the maliciously secure mixed-protocol MPC framework based on these share conversion protocols. This is the first maliciously secure mixed-protocol MPC framework relying on the standard model, providing a higher security guarantee than all the previous works in the literature. Also, this is the first general mixed-protocol MPC framework for n parties in the malicious model, in comparison to previous works that either only support fixed number of parties in the malicious model, or only handle limited types of share conversions. Furthermore, we provide the theoretical analysis of the computation and communication costs for the six types of share conversion protocols, as an important reference for future developers, who intend to implement some complex computation task by following this mixed-protocol MPC framework with malicious security.