Structural health monitoring has been the focus of recent developments in vibration-based assessment and, more recently, in the scope of the internet of things as measurement and computation become distributed. Data has become abundant even though the transmission is not always feasible, especially in remote applications. It is thus essential to devise data-driven model workflows that ensure the best compromise between model accuracy for condition assessment and the computational resources needed for embedded solutions. This topic has not been widely used in the context of vibration-based measurements. In this context, the present research proposes two approaches for two applications, a static and a rotating one. In case one, a modeling workflow capable of reducing the dimension of autoregressive model features using principal component analysis and classifying this data using some of the main machine learning techniques such as logistic regression, support vector machines, decision tree classifier, k-nearest neighborhood and random forest classifier was proposed. The three-story building example was used to demonstrate the method s effectiveness, together with ways to assess the best compromise between accuracy and model size. In case two, a test rig composed of rotating inertias and slender connecting rods is used, and the monitoring solution was tested in an embedded GPU-based platform. The models implemented to effectively distinguish between different friction states were principal component analysis, deep autoencoder and artificial neural networks. Shallow models perform better concerning running time and accuracy in detecting faulty conditions.