There are many ways to add some details

But lets consider your dilemma. The choice depends on your purpose. I'm not an expert, but my conclusions are:

Separate object

Advantages

• is easier to animate, it has its own axes etc, etc
• easier to implement, you don't have to break existing low poly edge flow, you can just attach it
• it may be even floating in front of object, and if you render it with low dimension there might not be visible difference (check this)

Disadvantages

• in case you want to render something when your camera is quite close to an object there could be some undesirable artefacts. Finally it depends on the tool this model will be used by (game engine, just renderer or something else)

Embedded geometry

Advantages

• precise control over the edge flow
• no artefacts in case everything is properly done

Disadvantages

• you model are highly likely to turn into high poly. How many polygons can you afford?

Normal map or bump

Create two objects (low and high poly) instead of one and then prepare normal map (just search through the internet how to do that, there are a lot of tutorials)

Advantages

• no redundant geometry at all! be truly low poly!
• quite good looking on the final render, ideal for games

Disadvantages

• requires more time to prepare, knowledge of editors such as Photoshop, knowledge of texturing and might be something else

And something else

• this is not real geometry, it will be visible only on the final render. Shaders will create lights and shadows in such way that imitates presence of geometry. What is it, advantage or not? You can decide according to what you are trying to achieve

Displacement modifier

Well, in case it is not necessary to keep you object in low poly you can tesselate desired polygons as many times as you need and then apply Displacement map. In that case you get extruded polygons in 5 minutes time without any modelling

Advantages

• extremely fast to implement and achieve some results, almost no modelling at all

Disadvantages

• forget about low poly
• requires prepared mask for displacement

And something else

• some game engines are clever enough to process displacement maps on fly, and also DirectX11 has tessalation features. Your models

So, what to choose is entirely up to you!

There are many ways to add some details

But lets consider your dilemma. The choice depends on your purpose. I'm not an expert, but my conclusions are:

Separate object

Advantages

• is easier to animate, it has its own axes etc, etc
• easier to implement, you don't have to break existing low poly edge flow, you can just attach it
• it may be even floating in front of object, and if you render it with low dimension there might not be visible difference (check this)

Disadvantages

• in case you want to render something when your camera is quite close to an object there could be some undesirable artefacts. Finally it depends on the tool this model will be used by (game engine, just renderer or something else)

Embedded geometry

Advantages

• precise control over the edge flow
• no artefacts in case everything is properly done

Disadvantages

• you model are highly likely to turn into high poly. How many polygons can you afford?

Normal map or bump

Create two objects (low and high poly) instead of one and then prepare normal map (just search through the internet how to do that, there are a lot of tutorials)

Advantages

• no redundant geometry at all! be truly low poly!
• quite good looking on the final render, ideal for games

Disadvantages

• requires more time to prepare, knowledge of editors such as Photoshop, knowledge of texturing and might be something else

And something else

• this is not real geometry, it will be visible only on the final render. Shaders will create lights and shadows in such way that imitates presence of geometry. What is it, advantage or not? You can decide according to what you are trying to achieve

Displacement modifier

Well, in case it is not necessary to keep you object in low poly you can tesselate desired polygons as many times as you need and then apply Displacement map. In that case you get extruded polygons in 5 minutes time without any modelling

Advantages

• extremely fast to implement and achieve some results, almost no modelling at all

Disadvantages

• forget about low poly
• requires prepared mask for displacement

And something else

• some game engines are clever enough to process displacement maps on fly, and also DirectX11 has tessalation features. Your models will be low poly, so to say, delegate your job to someone else

So, what to choose is entirely up to you!