In this paper, we develop an analytical model for radio resource dimensioning for latency-critical services in 5G networks. URLLC (Ultra-Reliable Low Latency Communications) service is introduced in the 5G networks to respond to the requirements of critical applications such as self driving cars, industry 4.0, etc. Its stringent requirements in terms of latency and reliability are challenging to meet and are usually tackled by resource over-dimensioning. In this paper, we develop large deviation bounds for the outage probability i.e., the probability that the packet delay exceeds a given target. Our numerical applications show that the bounds are sufficiently tight for mastering over-dimensioning. We then develop a resource dimensioning framework based on the developed bounds and apply it to a large scale system level simulator. Our simulation results show that the developed model, when coupled with field-based radio condition distributions, allows achieving the reliability targets with acceptable cost in terms of resource consumption.