The technology of embedded systems is now going through a period of revolution with unique improvements in the field of machine learning. Embedded machine learning (EML) is employed in many applications such as healthcare, computer applications, speech recognition, robotics, and other domains. However, the embedded application machine learning techniques are limited in their implementation. Machine learning algorithms are known to need a lot of memory usage, and hence it is incompatible for embedded and mobile applications. Embedded and mobile computing need special algorithms and hardware optimizations to run these compute- and memory-intensive methods. Thus, this paper examines current research trends inside this boundary with a brief summary of the numerous computationally intensive machine learning algorithms (Hidden Markov models HMMs, K-Nearest Neighbor’s-k-NNs, Support Vector Machines—SVMs, Gaussian Mixture Model GMMs, and deep neural network techniques). And, also, different optimization methods are also explored to fit these computational and memory-intensive algorithms into embedded and mobile environments with limited resources. We also discuss how these algorithms are implemented in hardware accelerators, mobile devices, and microcontrollers in real-world applications. Finally, the presented work emphasizes the key takeaways of EML technology applications and promising research topics for future work.