Cloud applications that utilize the "Functions as a Service" (FaaS) model, have gained significant popularity. However, their stateless nature necessitates frequent interaction with an external data store, which can limit their performance. To address this issue, a system for FaaS platforms that is transparent, vertically and horizontally elastic, and distributed across worker nodes. It mitigates effectively the problem by leveraging two common sources of resource wastage (I) cloud tenants typically observe overprovision memory resources for their functions due to their non-trivial input-dependence, and (ii) FaaS providers keep function sandboxes active for several minutes to avoid cold starts. By utilizing machine learning models tailored to different types of function input data including data mining, multimedia applications and gaming. We employed a novel technique that classifies the data in Azure Storage Blobs in hot and cold. Hot data is used frequently and cold data is not frequently required. There are no straight forward annotations or features that directly label the hot items. We have to examine the attributes of each data item and segregate them into classes. We employed a Naïve Bayes classifier to mark the data items with improvisation of Sine- Cosine optimization for hyper parameter tuning. This methodology significantly boosts the accuracy from 57.44% to 94.34%.