Due to the complexity of the current software, the measurement processes become crucial activities. However, due to the quantity of aspects to be measured and thus the big amount of data to manipulate, the software measurement plans are heavy to manage. It leads to very complex measurement plans engendering eventual losses of time and performance. The main objective of our paper is the improvement of the measurement plans by making the metrics use more flexible. This is an important requirements for the project managers. This allows to tackle specific useful metrics in avoiding measures that are not always relevant during an identified measured period of time. We propose to analyze and classify the measurements at runtime using a learning approach (Support Vector Machine, SVM) in order to define the relevant metrics that should be used at a specific time t. We designed a suggestion process that selects metrics from a current measurement plan or reorient (suggest) that measurement plan by proposing to execute other metrics. We implemented our framework on an efficient platform and successfully ran several experiments that we discuss and comment