The alpha channel tells the story. The first stage's RGB is not premultiplied, and the alpha channel set to black -- making the particle a complete addition mask. Note that I've chosen the particle's color to be a shade of orange that still has some color in the blue channel -- otherwise adding these particles on top of each other will give pure yellow, instead of the white we're aiming at. The particle's ADD mask is built to soften the edge of the flame in the ignition stage.
The second stage is similar to the first stage, except the add mask is softer -- the color added is darker, and the add factor is lower. This serves as a good transition between the bright ignition and the dark soot. The texture is not premultiplied, although it may seem that way at first glance.
The smoke is very simple -- it is a grey, alpha premultiplied particle, which only occludes the scene. We use a grey that is lighter than the shade of smoke we want, since we'll also be fading the particle to transparent black as its end-of-life approaches.
The emitter is a simple point emitter that emits 40-50 particles per second. It sets the particles lifetime and speed (height of the flame), its speed direction (flame spread), and its grow rate (flame's "body"). In addition, each particle is created with a slightly different semitransparent white "tint" (flame intensity) and finally, as mentioned before, each particle is tinted to transparent black toward the end of its life.
With a little trial-and-error we calibrate the emitter to achieve the final result:

There is plenty of room for modifications -- for example, by changing the time each particle spends in each stage we can get more intense ignition, or a longer smoke trail. We could also apply physical modifiers to each particle, adding wind or gravity for instance.
Conclusion
With the current product proliferation in the casual mobile gaming market, focusing on a single hit is a risk. Instead, publishing many titles quickly provide risk distribution and operational latitude to respond to the market's feedback, and experiment with different types of games and user experiences.
To achieve that, game development cycles need to shorten, in time and in cost. We focused on a factor that can affect both significantly -- graphics. Unless the gameplay explicitly requires 3D environment, I suggested choosing 2D, which is usually lighter on artist and developer resources.
As for user experience, we've seen a basic example that shows that a small amount of work and very basic 2D constructs can be used to generate elaborate graphic presentation.
But 2D effects aren't limited to the method I've presented. Even though it's in use all over the industry, over great range of complexities, this was just an example of what can be done with very basic building blocks. Many other effects can be achieved with the same facilities, giving developers great creative freedom in developing 2D games and users the visual experience they expect, even without impressive 3D environments and characters.