Mixed numerology was introduced in the physical layer of the 5G New Radio standard to give it the flexibility needed to serve diverse use cases and services such as enhanced mobile broadband, massive machine-type communications, and ultra-reliable low latency communications. It is inherently well-suited for supporting network slicing. The subcarriers of different numerologies have different bandwidths and symbol durations. As a result, they interfere with each other despite being centered at different frequencies. We have developed a novel analysis of the inter-numerology interference (INI) power that arises in these systems. Our comprehensive model accounts for the wideband, time-varying nature of the channels seen by 5G deployments, and the presence of phase noise, which is a serious impediment at millimeter-wave frequencies. It applies to the entire family of numerologies of 5G NR, and accounts for partial subcarrier loading, guard bands, and non-line-of-sight and line-of-sight channels. We also propose a novel statistical square root power allocation scheme that exploits the variation in the average INI powers across the subcarriers to mitigate the impact of INI.