Function as a Service (FaaS) has emerged as a prevalent cloud computing paradigm, favored by its ease of deployment, rapid elasticity, and pay-as-you-go billing. However, on-demand resource management of FaaS often introduces the cold start problem, which can increase waiting time and degrade performance. Inspired by the divide-and-conquer idea, this paper introduces HybridFP, a Hybrid Function Provision method designed to effectively and economically mitigate cold starts. We first partition serverless functions into model-worthy and model-unworthy functions in view of prediction profit. Subsequently, HybridFP integrates an invocation predictor and a pattern matcher to guide function provision, enabling timely function prewarming and instance eviction. Experimental results conducted on two industrial datasets demonstrate that HybridFP achieves performance on par with the state-of-the-art in mitigating cold starts, while reducing memory waste by 78.29% and 88.6%, respectively.