I need to sense temperature for a lithium battery pack and a TO-220 package transistor. Would an NTC with voltage divider be ok or is it better to use a sensor like the LM36?
Note:
There's nothing wrong with an NTC thermistor, it's as 'proper' as anything. Silicon sensors maybe be better or worse than in some regards. As may RTDs or thermocouples or other more exotic sensors.
As well as accuracy and stability (which are usually quite good for an NTC thermistor, at least over a modest temperature range) consider how well you can couple the sensor to what you are trying to measure. Most temperature sensors do an excellent job of measuring their own temperature, however that reading may lag behind (due to poor heat transfer and large heat capacity- especially an issue with still air temperature measurements) or have a large offset to what you are actually intending to measure due to heat loss down the leads, self-heating etc.
Thermistors are cheaply available with very tight tolerance thanks to consumer mass markets. Something like an LM35 or LM335 is accurate to around +/-1°C. A thermistor could be more like +/-0.2°C - 5x better for maybe 1/20 the cost.
For a battery temperature measurement, something like a 10kΩ NTC thermistor is an appropriate solution and is the standard approach. It doesn't need to be a 1% type but you can use one if you like. If you'd prefer (for some reason) to avoid any analog electronics and care little about cost effectiveness, monolithic digital output sensors are available.
A NTC thermistor is actually a pretty good sensor if implemented well. I've done something similar to yours (keeping an eye to the temperature of a Li-Ion cell and adjusting charge current to avoid overheating, of course via MCU code). This works remarkably well and is really easy to implement, as you say, just a voltage divider (but keep an eye on the ADC impedance requirements).
In a different scenario I've used a similar 10K thermistor to measure the temperature rise of a LED PCB, with the thermistor attached to a decent DMM with a serial interface to a PC, and some python code to convert the resistance to temperature, this apparently measures temperature to sub °C accuracy. When left to rest, the apparent measurement noise is again in the <0.1°C range.
Would an NTC with voltage divider be ok or is it better to use a sensor like the LM36?
Choose the device that meets your accuracy requirements (data sheet information), is easiest to implement, is a decent price, is reliable and, is sourced from a recognized dealer in electronic components. Whatever device you choose it should have a decent data sheet and be made by a recognized manufacturer.
There is no abolute "proper" or "better" solution in electronics, those notions exist only in relationship with your spec.
Quantify what you mean by "acceptable", then choose accordingly.
It is no problem if accuracy differs a bit
What do you mean by "a bit"? 1°? 10°?, 0.1°?
Once you know that, you can compute the accuracy of your temperature sensing circuit, and choose the solution that fits you precision requirement best, while still optimizing as many other parameters (cost, for example) as possible.
