Recent developments in hardware systems, sensors, and wireless communications technologies have caused an enormous growth in the number of devices with wireless connection capabilities. Most of these devices contain embedded systems that perform simple tasks such as (i) activity detection, e.g., in smartwatches, and health monitors, (ii) data collection and transmission, e.g., in remote sensors, and (iii) localization, e.g., in smart tags and industrial automation systems. However, such devices have limited battery capacity, limited bandwidth, and limited computational power to enable high mobility at low cost. Because of these limitations, embedded systems need (i) hardware-efficient and energy-efficient sensing pipelines that minimize the production cost and energy consumption, (ii) sample-efficient and complexity-efficient sensing pipelines that optimize radio-frequency (RF) spectrum utilization among many wireless devices, and (iii) cost-efficient technologies that enable accurate localization. This thesis proposes low-cost solutions to the concerns referred to above by introducing novel means of sensing and localization methods for embedded systems. In particular, to minimize the hardware complexity and energy consumption at low cost, we present a hardware-efficient and cost-efficient signal classification pipeline that extracts suitable features directly in the analog domain. To optimize utilization of the radio-frequency (RF) spectrum, we present a sample-efficient sensing pipeline that identifies unused resources in both frequency and space. To locate the devices using low-cost off-the-shelf wireless transceivers, we present novel approaches that enable accurate localization using channel state information (CSI) and deep neural networks (DNNs). For all of our solutions, we support our findings via theoretical analyses, simulations, and real-life experiments.