Time series sequences include a series of values recorded in specific time intervals which expose the functionality and behavior of data elements in a respective domain. Grouping of these elements based on the trend of their attribute values is challenging in an unsupervised learning environment. New representational structures are needed to explore the trend among unlabeled time series elements through clustering. We propose a fuzzy representational method and structure named fuzzylets, which can be used for unsupervised clustering of time series elements based on the trend of respective series. Fuzzylets provide a flexibility for various time series elements to discover their similarity based on the fuzzy membership of the trend existing among them. The fuzzylets are clustered using traditional hierarchical clustering and compared with other methods using the silhouette scores obtained from the clustering results. We performed an experimental analysis with fuzzylets on the electric energy dataset which contains the inflow and outflow of renewable energy in continental Europe for the tenure from 2012 to 2014 and UCR-2018 time series database containing 128 datasets. We compared fuzzylet based classification with six traditional methods. Fuzzylet based classification algorithms show better accuracy than others which reveals the importance of this novel time series primitives in time series feature learning.